Most AI ROI claims are estimates. Measure the hours a person stops spending on a named task, and check again a month later. Subtract the time spent checking the AI’s work, and you have a number you can trust.
Licenses don’t prove work got done
Vendors count what’s easy: seats, prompts, and minutes “saved” in a survey. An MIT report found most companies saw no measurable profit impact from their generative AI pilots. Yet individual workers said the tools made them more productive. The time saved was real, but nobody tracked where it went.
Measure one task at a time
You need one task and a stopwatch, not a dashboard.
- Name the task. Pick something specific and recurring, like prospect research or meeting prep, not “sales work”.
- Time it for a week. Write down how long it takes without AI.
- Hand it to AI. Set it up properly, then use AI for the task for a week.
- Time it again. Include time spent checking and fixing the AI’s work.
- Check after 30 days. Confirm you still use it, the savings held, and quality didn’t slip.
The gap between week one and day 30 is your real number.
A real number comes from measured work
A study of more than 5,000 customer support agents gives the best-known measured result. With an AI assistant, they resolved 14% more issues per hour on average. The least experienced agents improved by 34%.
Your number works the same way. Here is one person doing prospect research:
| Prospect research, one person | Hours a week |
|---|---|
| Before AI | 5 |
| With AI, including checking | 2 |
| Hours given back | 3, about 150 a year |
Multiply those hours by what the person’s time is worth, and you have an ROI figure you can defend.
Count the hidden costs
Reviewing the AI’s work takes time, especially at first. Errors that slip through cost more to fix later than to avoid. Someone has to set up and maintain the prompts and connections. Leave these out and your number runs high.
Peers tell you what’s realistic
Vendors publish only their best cases. Peers doing the same work will share what they measured, including what didn’t pay off. Members do that with numbers on Signals, the monthly PageLines call about what’s working in AI. To build this habit across a team, see AI leadership.
Bottom line: Real AI ROI is the hours a person stops spending on a named task, minus checking time, confirmed a month later. Measure one task at a time and count the hidden costs. PageLines members share these numbers on Signals every month. Learn about Signals.
Common questions
How do you measure the ROI of AI?
Time one named task for a week. Hand it to AI and time it again, including checking the AI's work. After 30 days, confirm the savings and quality held.
Why do most companies see no ROI from AI?
Many count licenses or usage, not work done. An MIT report found most companies saw no measurable profit from generative AI pilots, though workers reported time saved.
What are the hidden costs of AI?
Reviewing the AI's work, fixing errors that slip through, and maintaining prompts and connections. Skip them and your ROI number runs high.

